Yield-Curve Based Probability Forecasts of U.S. Recessions: Stability and Dynamics

Various papers indicate that the yield-curve has superior predictive power for U.S. recessions. However, there is controversial evidence on the stability of the predictive relationship and it has remained unclear how the persistence of the underlying binary recession indicator should be taken into account. We show that a yield-curve based probit model treating the binary recession series as a nonhomogeneous first-order Markov chain sufficiently captures the persistence of the U.S. business cycles and produces recession probability forecasts that outperform those based on a conventional static model. We obtain evidence for instability in the predictive content of the yield-curve that centers on a structural change in the early 1980s. We conclude that the simple dynamic model with parameters estimated using data after the breakpoint is likely to provide useful probability forecasts of U.S. recessions in the future.